github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-04-28 19:55:29 +00:00
Code Issues Packages Projects Releases Wiki Activity

add args.prefetch_num for benchmark

Browse Source
This commit is contained in:
genghaozhe 2024-05-25 14:55:50 +00:00
parent fba04e857b
commit b9269d962d
1 changed files with 4 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
examples/language/llama/benchmark.py
View File

@ -79,7 +79,7 @@ def main():
parser.add_argument("--custom-ckpt", action="store_true", help="Customize checkpoint", default=False) parser.add_argument("--custom-ckpt", action="store_true", help="Customize checkpoint", default=False)
parser.add_argument("--profile", action="store_true", help="Enable profiling", default=False) parser.add_argument("--profile", action="store_true", help="Enable profiling", default=False)
parser.add_argument("--disable-async-reduce", action="store_true", help="Customize checkpoint", default=False) parser.add_argument("--disable-async-reduce", action="store_true", help="Customize checkpoint", default=False)
parser.add_argument("--prefetch_num", type=int, default=0, help="chunk prefetch max number")
args = parser.parse_args() args = parser.parse_args()
colossalai.launch_from_torch() colossalai.launch_from_torch()
@ -114,7 +114,7 @@ def main():
extra_dp_size=args.extra_dp, extra_dp_size=args.extra_dp,
enable_fused_normalization=torch.cuda.is_available(), enable_fused_normalization=torch.cuda.is_available(),
enable_flash_attention=args.xformers, enable_flash_attention=args.xformers,
max_prefetch=10, max_prefetch=args.prefetch_num,
enable_async_reduce=not args.disable_async_reduce, enable_async_reduce=not args.disable_async_reduce,
) )
elif args.plugin == "gemini_auto": elif args.plugin == "gemini_auto":
@ -125,6 +125,8 @@ def main():
tp_size=args.tp, tp_size=args.tp,
extra_dp_size=args.extra_dp, extra_dp_size=args.extra_dp,
enable_fused_normalization=torch.cuda.is_available(), enable_fused_normalization=torch.cuda.is_available(),
max_prefetch=args.prefetch_num,
enable_async_reduce=not args.disable_async_reduce,
enable_flash_attention=args.xformers, enable_flash_attention=args.xformers,
) )
elif args.plugin == "fsdp": elif args.plugin == "fsdp":